The CREO Project: Carbon dioxide Reduction through Emissions Optimisation and the Use of MATLAB

Transcription

1 The CREO Project: Carbon dioxide Reduction through Emissions Optimisation and the Use of MATLAB MATLAB Expo 2013 Bob Lygoe Technical Specialist, Calibration CAE and Optimisation Richard Sykes, Pete Dowell Calibration Systems Integration

2 The CREO Project: Overview 1. What is CREO? CO 2 Reduction through Emissions Optimisation A collaborative R&D project funded by the TSB 2. Who is involved? 3. What is the aim of the project? A 15% estimated reduction in CO2 within business and customer constraints. Previous 3 decades has seen engine efficiency deteriorate by an estimated 8% to meet increasingly stringent vehicle emissions legislation. 4. Impact? 500K engines per annum within 7 years. 3 million engines per annum globally within 15 years. Page 2

3 The CREO Project: Overview (cont d.) 5. How will CREO achieve its aim? re-design the engine and exhaust after-treatment as a complete system, meeting all the legislative, customer and business requirements while minimising the CO2 levels. 6. Specific goals? i. The use of novel exhaust after-treatment techniques. ii. The on-board generation and use of hydrogen. iii. The development and application of new emissions optimisation tools. Jaguar Land Rover led gasoline workstream. Ford led diesel workstream. + This work is concerned with the Ford path and will describe how MATLAB has been used to help develop an emissions optimisation process. Page 3

4 Emissions Optimisation Process scvsp Design Space Speed, Torque, EGR, Boost, SOI Define Problem Speed, Torque Collect DoE Data Data Collection Transient Engine Model Steady State Deviations Transient Airpath Model Calibration EGR, Boost, SOI Global DoE Emission Model FG Emissions Exhaust Flow / Temperature Aftertreatment Model Tailpipe Emission Model Fuel Economy NOx Particulates HC CO Noise Page 4

5 Emissions Optimisation Process scvsp Design Space Speed, Torque, EGR, Boost, SOI Define Problem Speed, Torque Collect DoE Data Data Collection Transient Engine Model Aftertreatment Model Optimise DoE Steady Design State Deviations Transient Airpath Model Calibration EGR, Boost, SOI Global DoE Emission Model Optimise Objectives Tailpipe Emission Model FG Emissions Exhaust Flow / Temperature Fuel Economy NOx Particulates HC CO Noise Page 5

6 Emissions Optimisation Process: vehicle simulation SCVSP Model-Based-Design is an important tool in efficiently delivering a range vehicles with an optimised powertrain maintaining minimum complexity Using a physics-based prediction over various legislative and customer driving cycles before a vehicle exists, provides the input data to modelbased optimisation of the engine. SAE paper: Page 6

7 Emissions Optimisation Process: DoE One factor at a time: Requires expert consultation. One result with narrow boundary constraints & targets. DoE process: Minimal testing with maximum information content e.g. 9 factors, 3 levels, quadratic polynomial model: Full factorial: 3 9 tests 2 months. D-optimal: 55 tests 5 hours. Can optimise over a range of boundary constraints and targets Fitting models to data enables optimisations. Optimise for multiple scenarios without additional testing. Optimal trade-off (Pareto front) analysis. Analysis of correlation between inputs & responses. Automated testing Quicker and more repeatable. Requires less interaction can run overnight. Page 7

8 Emissions Optimisation Process: DoE One factor at a time Start point DoE Collect DoE Data Initial constraint Initial constraint Page 8

9 Emissions Optimisation Process: DoE One factor at a time DoE Calculate Pareto front Chosen solution Constrained optimum Page 9

10 Emissions Optimisation Process: DoE One factor at a time DoE Non-optimal solution Re-optimised to meet constraint without additional testing Revised constraint Revised constraint Page 10

11 Emissions Optimisation Process: Local vs. Global DoE Local DoEs Point-by-point (time-weighted) cycle optimisation Can only evaluate model at discrete speed/load points Requires interpolation to get an accurate cycle prediction Requires accurate & repeatable dynamometer control Global DoEs Requires automated screening More efficient data collection Less points for the same speed/torque area Infinite speed/torque resolution optimise to calibration breakpoints Enables transient analysis & optimisation Requirement for after-treatment optimisation Enables engine scaling / dimensionless modelling Page 11

12 Torque [Nm] Emissions Optimisation Process: Airpath DoE Airpath DoE: Define airpath boundaries without any pre-existing knowledge of engine Run DoE using open loop control More stable EGR rate control Increased repeatability Capture relationship between limits and MAF/Boost/SOI Define safe boundary for global DoE Enables use of closed loop control during global DoE with less on-line screening (e.g. hit rate increased from 70% to 90% after airpath DoE) High Load % EGR % VGT SOI Low Load % EGR % VGT SOI Smoke T2 HC/CO EGR PR Smoke HC/CO Smoke T2 HC/CO EGR PR MAF Smoke HC/CO Speed [rev/min] Pmax Noise T3 SOI interaction MAF MAP MAP Turbo Turbo Turbo Page 12

13 Emissions Optimisation Process: Cycle Optimisation Engine-out emissions trade-off EU 6.1 NOx target More efficient NOx after-treatment => Lower CO2 Page 13

14 Emissions Optimisation Process: Cycle Optimisation (using Model-Based Calibration Toolbox) Min. weighted sum CO 2 Optimisation algorithm info. Sequence of optimisations with progressively tighter constraints Constraints Parameter starting values Operating points and weights Page 14

15 Emissions Optimisation Process: Cycle Optimisation (using Model-Based Calibration Toolbox) Optimisation Results Table Optimisation Results Plot Optimisation Results Graphs Page 15

16 Emissions Optimisation Process: Cycle Optimisation Next Steps: Currently experimenting with a multi-objective genetic algorithm to: Generate the Pareto-optimal front in one run. More exhaustively search the decision variable space. Reduced sensitivity (vs. gradient-based optimisers) to start position. Reduced sensitivity to getting stuck in local minima. This is a significant computing challenge due to the large number of decision variables and may require: Large populations. Parallel computing. Page 16

17 Emissions Optimisation Process: Transient Emissions WAVE-RT Transient Airpath Model ECU Calibration Empirical DoE-based Model Library Inputs Outputs Page 17

18 Engine Torque [Nm] Emissions Optimisation Process: Transient Torque vs. Vehicle Measured Model Time [s] Page 18

19 Emissions Optimisation Process: Aftertreatment Model Emissions After treatment model Chemical Kinetic models execution time is typically too large to be run in the loop with iterative engine optimization Empirical lookup models are utilized to allow rapid evaluation of engine and after treatment as a system. Engine Model Process - Thermal Modelling - NOx Conversion Modelling Outputs - Exhaust Temperature - Mass Flows - Gas Composition - Pressure Aftertreatment Model Outputs - Tailpipe - Emissions Page 19

20 Emissions Optimisation Process: Aftertreatment Model Output Page 20

21 Summary A model-based emissions optimisation process has been developed in the MATLAB/Simulink environment. This process has yielded fuel economy improvements as well as time/cost efficiencies. Furthermore, it enables: Full exploration of the design space for any fuel economy gains. What-If studies to explore design and calibration opportunities. Robustness studies using optimisation to explore the impact of real-world noise sources (wear, manufacturing, environment, customer usage). Further work to be done includes development of: Better correlation of emissions models. Optimisation capability for high-dimensional problems. Further process automation. User interfaces. Page 21